Broadband Train Wreck

By Gary Kim

"The broadband value chain is headed for a train wreck," many within the industry might argue. The reason is simply that incentives for network operators to support many of the bandwidth-intensive innovations planned by upstream industries and users are not clearly in place.

In fact, inability to capture value created by applications that ride on broadband perversely creates positive disincentives for network transport and access providers to put gobs of bandwidth into place. Flat-fee bandwidth pricing is one frequently noted culprit, as such business models provide little incentive for users to regulate their own behavzior in ways that match usage and revenue.

"Many operators have also proposed to respond to rising usage-based costs by extracting additional revenue from value-added services beyond basic access, such as voice-over-IP and IP-based television," says the Massachusetts Institute of Technology’s Broadband Working Group. But that won’t work, the MIT group says.

"We do not believe this response is adequate to solve the problem, for two reasons," says the working group. "First, we expect that operators’ revenues from value-added services will be insufficient to cover rising usage costs, because service revenues will be limited by competition from a growing set of third parties, and by legal or regulatory attention to any perceived constraints on such competition."

"Second, some bandwidth-intensive broadband innovations will not have an associated revenue-generating service, even though the innovations they represent will be valuable to users and upstream industries," the working group says.

It is important to note, the group says, that broadband traffic and user behavior differs from narrowband in a number of important ways. First, compared to narrowband usage, bandwidth demand is much more variable for any single user. Over the course of a day or a month, any single user’s consumption will show much more dynamic range in a broadband context, than in a narrowband context.

By definition, the number of users whose bandwidth consumption exhibits wider dynamic range increases as broadband becomes the normal way users connect.

Also, narrowband users regulated their usage simply because using bandwidth-intensive applications was a relatively unsatisfactory experience. The need to "log on" likewise limited total online time. Broadband tends to be more an "always on" experience, with obvious new demands on bandwidth.

Then, because more bandwidth is available, applications assume the existence of broadband and increase the amount of bandwidth applications consume.

Evidence from Korea, where broadband penetration is the highest in the world, suggests that mean traffic per user is rising rapidly. Aggregate network traffic nearly doubled

every year since 2001, while the number of subscribers grew at a much slower rate during that period, the working group notes.

Additionally, a small fraction of users generates a much larger proportion of the total traffic. Korea Telecom reports that five percent of users generate half the total network traffic. As a result, KT has announced it will move from flat rate to usage-based pricing.

In the Australian market, the same trend can be seen: bandwidth consumption is increasing at a nonlinear rate higher than addition of users would suggest.

So why don’t broadband access providers simply raise the flat rates? In large part because that solution doesn’t fit the actual usage patterns most users exhibit. There are lots of light users.

So how about tiered pricing based on peak rate pricing, much as other products, such as voice, have been priced? The problem, say providers, is that peak rate pricing doesn’t correspond to usage-based costs. And a small number of users imposing heavy load on the network is the problem, not "peak" usage in aggregate.

"Peak rate" pricing also discriminates against some applications that are highly "bursty" but not necessarily bandwidth intensive in other ways. In other cases low latency is a huge requirement, though bandwidth isn’t (multiplayer games, for example).

A more logical approach then might be variable pricing based on monthly usage. To do so, providers will need better ways of estimating any particular’s usage in ways that make sense to end users, so the right plan can be assigned. The problem is that "gigabytes of usage" never make as much intuitive sense to end users as "minutes talked" or "time online."

The other thing is that users can find they are consuming bandwidth without knowing it. Networked applications can impose usage that users are not aware of. And there is the problem of malware, besides.

Finally, when traffic is sent and received imposes variable actual costs on a network. Off-peak packets sent or received at 3 a.m. local time do not tax a network, or impose costs, equivalent to packets sent and received during the peak weekday evening hours.

"Brute force" solutions, such as throttling back traffic after some threshold has been reached, are not as good as sophisticated ways of matching prices to behavior.

Of course, there are other possibilities, such as charging for applications running over access pipes, in the same way "voice" or "cable TV" has been paid for. Competitive forces, of course, limit the potential amount of money such an approach can generate. In the voice arena, of course, pricing above marginal cost might be difficult.

Nor can service providers expect to be able to "degrade" or otherwise impair the performance of competing applications, a tack that theoretically would restore some of the value of a walled garden service.

In other cases applications may not have a clear revenue model in any case. Blogs, podcasting and peer to peer applications provide examples of that.

Put simply, revenues and costs are not in balance if one assumes a world where lots of video will be downloaded and uploaded from mass market customers, paying anything close to current prices. Taking a look at video consumption in the U.K. market, IP Development Network founder Jeremy Penston.

"If today’s TV habits were transposed onto the Internet at 1080 progressive, this traffic would cost the ISP £51.45 per month per subscriber," says Penston. "People are not going to spend £50 a month to receive television pictures that are available today on free-to-air, so either the ISP takes a bath and drowns or they block this traffic."

The conclusion is inescapable: Internet Service Providers will start blocking video and other content and hold pricing and monthly usage quotas where they are, or users and content providers will have to start paying more so the additional bandwidth can be acquired and paid for.

That suggests nothing is not some sort of "tiered" access based on bandwidth consumption, much as business customers pay less for sub-T1 bandwidth, more for T1s and yet more for DS-3 levels of bandwidth. Customers who really want to view a lot of video might wind up on a different access plan charged at a higher rate.

Evidence so far suggests the need for such tiering will grow as use of Web-delivered video grows. Already, says Ellacoya, the 24 percent of network users consuming video are responsible for 92 percent of the bandwidth consumption. The 56 percent of users who send and receive instant messages consume just one percent of network bandwidth.

All Web surfing and email activities consume just seven percent of network bandwidth, though 65 percent of active users do so. The point is that some useful applications won’t have any clear revenue stream to tax.

Raw increases in bandwidth could provide a solution if the networks continue to become so efficient that even increases in usage do not overwhelm the business model. Of course, video makes that assumption tenuous.

The other thing is that if targeted and personal advertising emerges as the business model for much video, the very complexity of such operations defeats the expectation of lower per-bit costs as networks are upgraded.

Today’s most commonly proposed solutions for the broadband monetization problem, in particular higher flat fees, revenue models based on value-added services, and cost-reducing network engineering techniques may be part of the answer, but are unlikely to prove sufficient by themselves, the working group says.

Good solutions to this problem need to align the incentives of network operators and upstream stakeholders, for example by enabling monetization of usage that imposes costs on
providers. IP